AskTheWay / dsh-auto-memory
git clone https://github.com/AskTheWay/dsh-auto-memory.git
v0.9.0 Β· 4d ago
Claude Code-style auto-memory plugin for DeepSeek Harness (dsh): typed memory files + MEMORY.md index auto-injected into the system prompt. File-only, no external services.
+0
stars, 24h Β· #30,309
+~0
stars, 7 days Β· #30,784
+111
stars, 30 days Β· #4,645
101
β
total
7
forks Β· 3 open issues
Star history (as recorded here)
LanguageTypeScript
LicenseMIT
Created2026-09-22 (1mo ago)
Last push2d ago
Watchers1
OwnerUser
GitHub topics
agent-harnessagent-memoryai-agentai-toolsauto-memoryclaude-codecoding-agentcontext-engineeringdeepseekdeepseek-harnessdshdsh-pluginllmllm-agentmemorymemory-systemnpm-packagepersistent-memoryplugintypescript
Links
README excerpt
# dsh-auto-memory English | δΈζ > ### Your dsh agent forgets everything you tell it. Every. Single. Session. > **Fix it with one command.** Claude Code-style persistent memory for > DeepSeek Harness β native, > zero servers, zero embeddings, zero setup. [code] Say *"Remember: I'm a Python backend engineer preparing for interviews"* today β open a brand-new session tomorrow, ask *"what do you know about me?"*, and it **remembers**. --- ## Highlights **0.7 β update-replace semantics**: consolidation updates facts in place instead of piling up near-duplicates, with a matching core shared between product and eval harness. Verified against the official desktop client (0.2.0-rc.2). Full history: CHANGELOG.md Β· Releases ## Claude Code has this. dsh didn't. Now it does. DeepSeek Harness is the hottest open agent harness on GitHub right now β models, tools, sandboxes, everything is a plugin. But it ships with **no memory subsystem at all**. The official answer is three *default-off* MCP configs to third-party servers, which the official docs themselves qualify: not auto-injected, no forgetting policy, substring-only search. Your agent has amnesia by design. `dsh-auto-memory` closes that gap natively: | | MCP bridge approach | **dsh-auto-memory** | |---|---|---| | Memories injected into **every** system prompt, automatically | β | β (zero tokens when empty) | | Typed memories: user / feedback / project / reference | β | β | | Workspace + user scope layers β no cross-project leakage | β | β | | Crash & concurrency safety (cross-process locks, orphan recovery) | β | β | | External services / databases / embeddings required | βββ | **none β just plain Markdown files** | Memories are ordinary files under `$DSH_HOME/memory/` β hand-editable, grep-able, git-friendly, yours. ## One minute to feel it [code] Or for real, in a chat: tell your agent things worth remembering. The model calls `memory_write` / `memory_read` / `memory_list` / `memory_delete`, following Claude Code's write discipline: **dedupe-and-update over piling up**, absolute dates only, `[[name]]` cross-links, `feedback` memories carry **Why:** / **How to apply:** lines. ## What the model actually sees Every request, one system-prompt section (order 4000) carries the index β re-evaluated per step, byte-budgeted, and **gone entirely when the store is empty**: [code] Chinese titles, YAML frontmatter, one file per memory β exactly the Claude Code `MEMORY.md` model, rebuilt natively on dsh's prompt-assembly pipeline. ## Hardened before first release This plugin survived a **12-agent adversarial code review** (680k tokens of source-level scrutiny) before v0.1.0. Five production-grade traps were caught and fixed β with regression tests β including two that would have been field incidents: - **The NTFS silent destroyer**: a memory named `memory` collides with `MEMORY.md` on case-insensitive filesystems β the write *succeeds* while destroying the record. Blocked by a reserved-name guard. - **The pois